Abstract: Aiming at the problem of predicting the remaining service life of proton exchange membrane fuel cell (PEMFC), this paper proposes a fusion prediction method based on the joint iteration of temporal convolutional neural network (TCN) and adaptive unscented Kalman filter (AUKF). This method first uses TCN for short-term prediction and uses Bayesian algorithm to calculate fusion weights. Then the discrete wavelet transform is used to decompose the aging data into fluctuation trends and aging trends. The fluctuation trend is predicted in a long-term iteration based on TCN. The aging trend is predicted in a long-term iteration based on the joint iteration of TCN and AUKF. The two trends are superimposed to obtain the long-term prediction results. Finally, fusion weights are used to fuse the long-term prediction results of multiple single PEMFCs. Based on the data verification of five single cells under two working conditions, the short-term prediction results show that TCN has high prediction accuracy. The long-term prediction results show that the fusion process reduces the impact of uneven aging between PEMFC cells and improves the accuracy of stack life prediction. stability.